How it works
The Brain operates by ingesting raw data, chunking it, and embedding it into an internal vector database (pgvector). This process enables:
- Grounded Answers: When a team member asks a question via the chat or Slack integration, the AI references specific chunks from the Brain rather than its general training data.
- Citations: Every fact provided by the AI includes citations back to the original source material. No more institutional knowledge walking out the door—if it’s in the data, the Brain knows it.
- Access Control: The Brain respects Role-Based Access Control (RBAC). A user searching the Brain or chatting with an agent can only retrieve chunks of data that they have permission to see.
Populating the Brain
There are three primary ways to fill the Organization Brain:Articles
Traditional wiki pages written natively in Corgtex. Best for high-level manifestos, policies, and narrative guides.
Sources (Files)
Flat files (PDFs, docs) uploaded directly into the pipeline.
Data Sources
Automated imports from external relational databases.
The Ingestion Pipeline
Whenever a source is uploaded or synced, it runs through the Corgtex ingestion workflow:- Extraction: Text is pulled from the raw file or database row.
- Chunking: The text is split into semantic chunks.
- Classification (Zero-LLM PII Regex): Chunks are analyzed locally to flag PII (Personally Identifiable Information) or confidential markers, ensuring sensitive data doesn’t leak into unrestricted workflows.
- Embedding: The chunks are converted into mathematical vectors and stored securely in the PostgreSQL database.